On October 26, 2026, Broadcom, Bank of America, Citi, Morgan Stanley, and Blackstone jointly announced that through a newly established special purpose vehicle (SPV), they had completed a total financing commitment of $63 billion to purchase Broadcom AI accelerator chips and data center networking equipment, which would then be leased to Claude developer Anthropic under a unified arrangement. Deliveries are expected to begin in 2027. This is one of the largest debt transactions to date related to AI compute buildout.
To understand the core of this transaction, one must first look at its financing layers. Bank of America, Citi, and Morgan Stanley arranged the first-lien secured financing, accounting for about 69% of the total, or approximately $43.5 billion; Blackstone exclusively committed to providing the second-lien secured financing, covering the remainder. The two tranches have a clear priority in the payment waterfall—in the event of default, first-lien lenders are repaid first, while Blackstone bears higher risk and expects higher returns. This tranched structure allows capital with different risk appetites to participate in the same transaction—a standard approach in traditional infrastructure financing, now transplanted to AI chip procurement.
Broadcom’s role in the overall structure is more complex than that of a simple chip seller. According to the announcement, Broadcom provides “Residual Value Support” to the first-lien lenders, with the support amount equal to the size of the first-lien financing. The economic implication is that if the value of these AI accelerators depreciates sharply during the lease term, lenders could face the risk that the collateral is insufficient to cover the debt; by assuming a cap on this depreciation risk, Broadcom effectively adds a layer of credit enhancement to the syndicated loan. This does not mean Broadcom is guaranteeing all of Anthropic’s debt, but it means that, to complete chip sales, Broadcom is willing to lend its credit support to the long-term value of its own products. For lenders, this is effectively the chipmaker itself certifying that the hardware will not rapidly depreciate to zero collateral value within a few years.
This transaction did not emerge out of thin air. In June this year, Apollo and Blackstone had already reached a prior $35 billion transaction with Broadcom, also through an SPV, to finance Anthropic’s compute expansion—in that transaction, Broadcom’s residual value support was capped at $29 billion. The new $63 billion transaction is a second-stage expansion on top of the prior framework; together, the two deals total nearly $100 billion. That figure alone shows that Anthropic’s compute buildout is far from a one-time purchase; it is a long-term path that requires a continuing relay of financing.
The timing of the announcement coincided with the same day Anthropic filed its S-1 registration statement with the U.S. Securities and Exchange Commission for a Nasdaq listing. The filing disclosed that Anthropic had revenue of $16.3 billion in the first half of 2026 and a net loss of $80.7 billion—but $72.4 billion of that was book losses from fair value changes on convertible notes and derivatives, rather than actual cash outflows; operating cash flow was positive at $3.9 billion over the same period, and its operating loss had narrowed sharply from $8.06 billion for full-year 2025 to $693 million. With compute financing and the IPO landing on the same day, Anthropic is effectively sending two signals to the market at once: it has a long-term intention to build an independent compute channel, and it is also seeking public endorsement from the capital markets.
From the perspective of the industry competitive landscape, Broadcom’s financing architecture directly serves its head-on competition with Nvidia. Through this transaction, Anthropic is purchasing Broadcom’s self-developed XPU AI accelerators, not Nvidia GPUs. Once Broadcom’s chip sales have a Wall Street-caliber financing absorption mechanism, they no longer depend solely on customers’ own willingness to spend capital—as long as the lease contract cash flows are reasonable, financial institutions are willing to advance the purchase funds. This “finance first, deliver later, then collect rent” model could, in theory, allow Broadcom’s chip orders to convert into actual shipments at a faster pace. In August this year, Nvidia also announced partnerships with six financial institutions, including Blackstone, planning to mobilize more than $500 billion for the AI sector to help customers finance purchases of Nvidia chips. The two chip giants are building their respective financial support networks at almost the same time, showing that pure technology competition is no longer enough; the financing capacity for compute procurement is becoming a new competitive dimension.
For the AI industry as a whole, the core significance of this transaction lies in how debt capital is repricing AI infrastructure risk. According to market data, a CoreWeave bond maturing in 2031 has a secondary-market yield of about 15.76%, and its five-year credit default swap (CDS) spread is about 1,148 basis points, while the five-year CDS spreads for Meta, Alphabet, and Microsoft over the same period are about 121, 83, and 55 basis points, respectively. The market demands vastly different risk premiums for AI infrastructure debt from different entities. The key support enabling Anthropic’s transaction to reach such a scale is that Broadcom, as the equipment manufacturer, provides a residual-value backstop, while the tranched priority structure dilutes risk, allowing bank lenders to treat these chips as assets with some collateral value rather than as pure credit risk exposure.
The dual-track push of the IPO and chip financing means Anthropic is simultaneously opening up both equity and debt funding channels. Once the IPO raises funds, the new cash can reduce reliance on debt financing and can also serve as an equity cushion for future larger-scale hardware purchases. The real test of Broadcom’s residual value support commitment will come between 2028 and 2030—by then, the pace of AI accelerator technology iteration will determine whether the market residual value of the hardware delivered in 2027 can support the financing valuation set at the outset. If this SPV plus tranched financing template proves replicable by other AI companies, the traditional “build your own data centers plus equity financing” path will come under direct pressure, and the financialization of compute may become a standard tool for the next wave of large AI companies to lock in infrastructure. Signals to watch include how alternative asset managers such as Blackstone price risk in similar structures, and how investors react specifically to the chip lease liability structure during Anthropic’s IPO roadshow.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接